Modelling the spatial distribution of a threatened butterfly: Impacts of scale and statistical technique [An article from: Landscape and Urban Planning]

This digital document is a journal article from Landscape and Urban Planning, published by Elsevier in 2007. The article is delivered in HTML format and is available in your Amazon.com Media Library immediately after purchase. You can view it with any web browser.

Description:
This paper compares the performance of five modelling methods in the prediction of a species distribution, using a data set describing the distribution of the threatened clouded apollo butterfly (Parnassius mnemosyne) in south-west Finland. The five statistical techniques included were: generalized linear models (GLM), generalized additive models (GAM), classification tree analysis (CTA), neural networks (ANN) and multiple adaptive regression splines (MARS). The accuracy of the models was examined at three spatial resolutions (1, 25 and 100ha) by area under the curve (AUC) and kappa statistics. All five modelling techniques had a relatively high discrimination capacity for the occurrence of clouded apollo. Classification tree analysis provided the least robust model performance. The differences between the other methods were small, although GAM and MARS provided marginally the best stability and performance. The most accurate models were developed for the resolutions of 1ha (highest AUC values) and 25ha (highest kappa values) and the least accurate models for the resolution of 100ha. Our work shows that modern modelling techniques can provide useful forecasts of species distributions in unsurveyed parts of landscapes and provide valuable contributions to conservation and management planning. However, the success of applying the new modelling tools can be influenced by the choice of statistical technique and especially of spatial resolution. In conclusion, small changes in the spatial scale may result in a clear decrease in the model performance and thus caution should be exercised when implementing the models and their predictions in practice.